Feature extraction and classification approach for the analysis of roadside video data

The research in this thesis focused on developing an automatic video analysis approach for roadside object detection and classification. It investigated the problem of detecting roadside objects in unstructured environments. The major problem during the detection of roadside objects is the extraction of appropriate features. Successful classification of objects heavily depends on good feature representation. Learning algorithms produce low accuracy if the feature representation is poor. Hence, the main objective of this research is to develop novel feature extraction and classification techniques which can perform an accurate detection that is fast enough for real-time application.